Step 1 Biostatistics: A Practical High-Yield Review
Biostatistics is a defined part of Step 1, not a bonus section or a promise of easy scores. The USMLE content outline places biostatistics, epidemiology, and population health in a 4–6% range. That makes it worth studying steadily, with an emphasis on applying concepts to a clinical stem rather than memorizing disconnected formulas.
The most durable approach is to draw the relationship in front of you, state what the denominator represents, and check whether the result makes clinical sense.
Start With the Question’s Axes
A 2x2 table is useful only after you identify what its rows and columns represent. For diagnostic testing, the axes are usually disease status and test result:
| Disease present | Disease absent | |
|---|---|---|
| Test positive | true positive | false positive |
| Test negative | false negative | true negative |
From this table:
- sensitivity = true positives / all with disease
- specificity = true negatives / all without disease
- positive predictive value = true positives / all positive tests
- negative predictive value = true negatives / all negative tests
Predictive values change with prevalence. Sensitivity and specificity describe the test in the studied population, but their practical use also depends on the clinical context and consequences of false results. “SnNOut” and “SpPin” are memory aids, not absolute rules: likelihood ratios, pretest probability, and context matter.
A 2x2 table can also be arranged with exposure and outcome as the axes. That is a different table. Relative risk requires risk among exposed and unexposed groups; it cannot be derived from a diagnostic-test table unless the axes actually describe exposure and outcome.
Treatment Effects: Use Absolute Risks First
For an adverse event, define the control event rate (CER) and experimental event rate (EER).
- absolute risk reduction (ARR) = CER − EER
- relative risk (RR) = EER / CER
- relative risk reduction (RRR) = 1 − RR
- number needed to treat (NNT) = 1 / ARR, when ARR is expressed as a proportion
Round NNT up to the next whole patient and state the time horizon. For an adverse effect that is more common with treatment, use absolute risk increase (ARI) = EER − CER and number needed to harm (NNH) = 1 / ARI. Always check which group has the larger event rate before naming the measure.
Study Design Shapes What You Can Infer
A cohort study begins with exposure and can estimate incidence and relative risk. A case-control study begins with outcome and commonly uses an odds ratio because investigators choose the number of cases and controls. A cross-sectional study measures a snapshot and can estimate prevalence. Randomization can reduce confounding, but loss to follow-up, nonadherence, and measurement problems can still affect an experiment.
When a question asks about causation, ask whether the design establishes temporality, whether comparison groups are credible, and what bias or confounding could still explain the association.
Bias Is a Mechanism, Not a Vocabulary List
Try to name what changed in the data-generating process:
- selection bias: the included sample differs systematically from the target population
- recall bias: participants remember a prior exposure differently because of their outcome
- lead-time bias: earlier detection appears to lengthen survival time without changing time of death
- confounding: a third factor is associated with both exposure and outcome and distorts the comparison
- observer or measurement bias: how an outcome is measured differs between groups
A clinical scenario matters more than a label. An apparent survival benefit after a screening program may be lead-time bias if patients are diagnosed earlier but do not live longer from biological disease onset.
Read P Values, Confidence Intervals, and Power Precisely
A p value is the probability of observing results at least as extreme as the data, assuming the null hypothesis and model assumptions are true. It is not the probability that the null hypothesis is true, and it does not measure clinical importance.
A usual two-sided confidence interval that excludes the null value is consistent with statistical significance at the corresponding level. The null is 1 for ratios such as RR and odds ratio, and 0 for differences such as risk difference or mean difference. Confidence intervals also show precision: a wide interval may be compatible with meaningful benefit and meaningful harm.
Power is the chance of detecting a specified true effect under a chosen model. It depends on sample size, expected effect size, variability, and alpha—not sample size alone. A negative study can be uninformative when it is underpowered.
Intention-to-Treat and Per-Protocol Answer Different Questions
An intention-to-treat analysis keeps participants in their randomized groups. It preserves the benefit of randomization, although missing data and how it is handled still matter. A per-protocol analysis focuses on adherence to the assigned protocol. It may be useful for a different question, but post-randomization differences in who adheres can introduce bias. It is not simply “people who completed the study,” and it is not automatically more or less conservative.
A Study Routine That Builds Transfer
Use a modest, repeatable cycle instead of promising yourself a fixed number of hours or questions:
- Draw a diagnostic or exposure-outcome table before calculating.
- Solve mixed questions and write the denominator in words.
- Review each wrong answer for the interpretation error, not only the missed formula.
- Revisit bias and study-design questions with new clinical scenarios.
- Return to concepts after a few days so recall, not recognition, does the work.
If a formula feels fragile, make up a small example with 100 patients and calculate the counts. This often reveals an inverted denominator or a risk difference expressed as a percent instead of a proportion.
For a broader study plan, see how to build a Step 1 schedule. If you want help identifying a pattern in practice questions, book a study-planning consult.